AMDGPU: Run AMDGPUCodeGenPrepare after inlining

[arsenm]

Summary

With leaf functions, this makes nonsensical decisions
based on the uniformity of the arguments.

[• tstellarAMD]

LGTM.

[arsenm]

r293525

lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

public:
 
   GCNTargetMachine &getGCNTargetMachine() const {
     return getTM<GCNTargetMachine>();
   }
 
   ScheduleDAGInstrs *
   createMachineScheduler(MachineSchedContext *C) const override;
 
-  void addIRPasses() override;
   bool addPreISel() override;
   void addMachineSSAOptimization() override;
   bool addInstSelector() override;
 #ifdef LLVM_BUILD_GLOBAL_ISEL
   bool addIRTranslator() override;
   bool addLegalizeMachineIR() override;
   bool addRegBankSelect() override;
   bool addGlobalInstructionSelect() override;
void AMDGPUPassConfig::addIRPasses() {
   addPass(createAlwaysInlinerLegacyPass());
   // We need to add the barrier noop pass, otherwise adding the function
   // inlining pass will cause all of the PassConfigs passes to be run
   // one function at a time, which means if we have a nodule with two
   // functions, then we will generate code for the first function
   // without ever running any passes on the second.
   addPass(createBarrierNoopPass());
 
+  const AMDGPUTargetMachine &TM = getAMDGPUTargetMachine();
+
+  if (TM.getTargetTriple().getArch() == Triple::amdgcn) {
+    // TODO: May want to move later or split into an early and late one.
+
+    addPass(createAMDGPUCodeGenPreparePass(
+              static_cast<const GCNTargetMachine *>(&TM)));
+  }
+
   // Handle uses of OpenCL image2d_t, image3d_t and sampler_t arguments.
   addPass(createAMDGPUOpenCLImageTypeLoweringPass());
 
-  const AMDGPUTargetMachine &TM = getAMDGPUTargetMachine();
   if (TM.getOptLevel() > CodeGenOpt::None) {
     addPass(createAMDGPUPromoteAlloca(&TM));
 
     if (EnableSROA)
       addPass(createSROAPass());
 
     addStraightLineScalarOptimizationPasses();
   }
void GCNPassConfig::addMachineSSAOptimization() {
   // instructions leftover after the operands are folded as well.
   //
   // XXX - Can we get away without running DeadMachineInstructionElim again?
   addPass(&SIFoldOperandsID);
   addPass(&DeadMachineInstructionElimID);
   addPass(&SILoadStoreOptimizerID);
 }
 
-void GCNPassConfig::addIRPasses() {
-  // TODO: May want to move later or split into an early and late one.
-  addPass(createAMDGPUCodeGenPreparePass(&getGCNTargetMachine()));
-
-  AMDGPUPassConfig::addIRPasses();
-}
-
 bool GCNPassConfig::addInstSelector() {
   AMDGPUPassConfig::addInstSelector();
   addPass(createSILowerI1CopiesPass());
   addPass(&SIFixSGPRCopiesID);
   return false;
 }
 
 #ifdef LLVM_BUILD_GLOBAL_ISEL